BetaTPred3: In silico platform for predicting and initiating β-turns in a protein at desired locations

Welcome to the official repository for BetaTPred3, a comprehensive in silico platform for analyzing, predicting, and designing β-turns in proteins. This resource is designed to support researchers in structural biology, protein engineering, and computational drug discovery. Web Server: https://webs.iiitd.edu.in/raghava/betatpred3/ Algorithm Details: http://crdd.osdd.net/raghava/betatpred3/algo.html Citation Singh, H., Singh, S., & Raghava, G. P. S. (2015). In silico platform for predicting and initiating β-turns in a protein at desired locations. Proteins: Structure, Function, and Bioinformatics, 83(5), 910–921. https://doi.org/10.1002/prot.24783 About the Platform BetaTPred3 is an updated and significantly expanded in silico platform for the prediction and design of β-turns in protein sequences. β-turns are the most common type of non-repetitive secondary structures, constituting on average 25% of all residues in protein chains. They play key roles in protein folding, stability, and molecular recognition. This platform introduces a novel turn-level prediction approach — predicting the complete β-turn as a unit of four consecutive residues, rather than predicting individual residues. This is a fundamental methodological improvement over all prior methods. Data sources integrated include: Protein Data Bank (PDB) — ~20,000 chains analyzed for propensity calculation BT426 — classic benchmark dataset (426 chains, 25% sequence identity cutoff) BT6376 — large, latest dataset of 6376 nonredundant protein chains Key Features Best Model Performance BetaTPred3 (Random Forest): 79% accuracy, 0.51 MCC on BT426 dataset Propensity-based method: 82% accuracy using ~18,000 PDB-derived propensity scores Comparable to or better than all existing methods on the BT426 benchmark Comprehensive Dataset ~20,000 PDB chains used for residue and pair propensity analysis BT426: standard benchmark for comparison with prior methods BT6376: large nonredundant dataset for improved training and validation Rich Feature Set (Random Forest Model) Binary profile of tetrapeptide Evolutionary information via PSSM (Position-Specific Scoring Matrix) profile PSIPRED-predicted secondary structure of the tetrapeptide Multiple Prediction Approaches Random Forest (BetaTPred3) — best overall predictor Propensity-based method — uses position-specific scores from ~18,000 PDB chains Turn-level prediction method — predicts the complete β-turn, not individual residues β-turn type prediction — nine types classified by dihedral angles (I, I', II, II', VIII, VIa1, VIa2, VIb, IV)